Web application to predict player ratings in the EA Sports FIFA 19 game

Frontend

Created using Angular 7 following the MVC design principle

Backend

Uses Flask to create API endpoints and Keras to create a regression neural network based on the 6 high-level attributes of a player:

• Pace: how fast a player is
• Shot: how good a player is at shooting
• Pass: how good a player is at passing
• Dribble: how good a player is at dribbling
• Defense: how good a player is at defending
• Physical: how aggresive, strong, etc a player is

The neural network is hosted using TensorFlow Serving

Development

1. Run npm i in the client directory to install all packages required for the frontend
2. In the server directory, install venv. Then, run python3 -m venv fut-env . This creates a virtual environment that will be used to install packages local to this project.
3. After running the above, run source fut-env/bin/activate. This activates the virtual environment. Next, run make install. This installs all the packages needed by the server
4. Start the jupyter server by running jupyter notebook. Open up the fut notebook and run all cells. This will fetch the data from the EA Sports FIFA endpoint, preprocess each JSON object, create the regression model, trains/evaluates it, and exports the trained model in HDF5 format.
5. Run make export. This converts the HDF5 model into a TensorFlow model graph, which can be used for serving.
6. To host the tensorflow server, run tensorflow_model_server --port=9000 --model_name=fut --model_base_path=$(pwd)/model/ in the server directory. This hosts the server on 0.0.0.0:9000
7. To start the frontend app, run npm start from the fut-client direcdtory
8. To start the backend server, run make run from the server directory. This starts the flask server on localhost:5000